fix: preserve pinned versions in DESCRIPTION and accept several dirs in dir.r (#139, #140)

[VincentGuyader]

Fixes two ways a version constraint written by hand in DESCRIPTION could vanish on att_amend_desc().

#140 - a hand-set version must never silently disappear

When a package was listed under two types (a pinned Imports and a bare Suggests, for instance), att_to_desc_from_is() rebuilt the dependency table by joining the scan result with deps_orig on package only. That multiplied the rows, and the later !duplicated() kept whichever row came first, so the surviving version depended on the order of the rows in DESCRIPTION. A pinned constraint could be reset to * with no message.

The fix collapses deps_orig to one row per package before the join, ordering so that an explicit constraint beats * and, on a tie, Imports beats Suggests. The hand-set version is now kept, deterministically, and when both an Imports and a Suggests row carry a version the Imports one wins. Type resolution itself (a package used from both an Imports and a Suggests location lands in Imports) was already handled by the scan and is unchanged.

#139 - dir.r now accepts several directories

att_amend_desc(dir.r = c("R", "inst")) raised the condition has length > 1: dir.r was filtered as a vector but then tested as a scalar with if (dir.r != ""). Changed to if (!identical(dir.r, "")) at both sites. The extractors (att_from_rscripts, att_from_examples) already vectorised over multiple directories. This lets you declare dependencies used only outside the default scan, such as a deployment entry point under inst/, so their pins are then preserved by the #140 fix.

Tests

New test-att_amend_desc_version_pins.R, written red-first:

• pin preserved when the package also sits under another type, including when the scan moves it to Suggests;
• the Imports-side version wins when both types carry different explicit constraints;
• an Imports/Suggests conflict resolves to Imports with its version;
• dir.r = c("R", "inst") no longer errors and a dependency used only in inst/ is detected.

Full suite green (77 tests). R CMD check --as-cran on the built tarball: 0 error, 0 warning, 0 relevant note. Submitted to win-builder R-devel.

Closes #139
Closes #140

[Copilot]

Pull request overview

This pull request improves {attachment}’s att_amend_desc() dependency rewriting so that maintainer-authored version constraints in DESCRIPTION are preserved deterministically, and extends dir.r to accept multiple directories so additional R-script locations (e.g. inst/) can be scanned.

Changes:

• Preserve hand-set version pins when a package appears multiple times in DESCRIPTION under different dependency types by collapsing original version rows before joining.
• Allow dir.r to be a character vector of directories (and document the behavior).
• Add focused regression tests for version pin preservation and multi-dir dir.r, and update NEWS/man pages.

Reviewed changes

Copilot reviewed 3 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/testthat/test-att_amend_desc_version_pins.R	Adds regression tests covering pinned-version preservation and multi-directory dir.r.
R/att_to_description.R	Fixes duplicate-package version-pin loss during DESCRIPTION rewrite; updates dir.r handling and roxygen docs.
NEWS.md	Documents the dir.r multi-directory support and the version-pin preservation fix.
man/attachment-deprecated.Rd	Updates generated documentation to reflect dir.r now accepting multiple directories.
man/att_amend_desc.Rd	Updates generated documentation to reflect dir.r now accepting multiple directories.

Files not reviewed (2)

• man/att_amend_desc.Rd: Generated file
• man/attachment-deprecated.Rd: Generated file

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[VincentGuyader]

Fixed in b3fcf9a. Also corrected two further occurrences of the same singular slip in this file (the test-name and the fakepkg comment), so Imports/Suggests are now written with their canonical plural field names throughout.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 7 changed files in this pull request and generated no new comments.

Files not reviewed (2)

• man/att_amend_desc.Rd: Generated file
• man/attachment-deprecated.Rd: Generated file

Suppressed comments (1)

R/att_to_description.R:379

• The tie-break comment says to prefer Imports over Suggests, but order(..., deps_orig$type) uses lexicographic ordering across all dependency types present in deps_orig (e.g., Depends/Enhances/LinkingTo), which can unintentionally take precedence over Imports/Suggests when the same package appears under other types with a different explicit constraint. Consider implementing an explicit type priority (e.g., Imports first, then Suggests, then everything else) and also treating NA versions consistently (since deps_orig$version == \"*\" can yield NA, affecting ordering).

  # Collapse the original versions to one row per package before the join.
  # A package listed under two types (a pinned Imports and a bare Suggests, say)
  # otherwise multiplies rows in the merge, and the later de-duplication would
  # keep whichever row happened to come first. Order so that an explicit
  # constraint beats "*", and Imports beats Suggests on a tie, so a version set
  # by hand in DESCRIPTION is never silently dropped.
  orig_versions <- deps_orig[
    order(deps_orig$version == "*", deps_orig$type),
    c("package", "version")]
  orig_versions <- orig_versions[!duplicated(orig_versions$package), ]